Instrument popular HPC libraries

[mofeing]

We should instrument the following popular libraries:

• Distributed
• Base.Threads
• MPI.jl
• Dagger.jl
• CUDA.jl
• oneAPI.jl
• AMDGPU.jl

[exaexa]

I'd target mostly Distributed and maybe Dagger. CUDA/AMDGPU already do a lot of ugly stuff with the source and I expect complications. For our usecases we might add DistributedData.jl

[mofeing]

I'd target mostly Distributed and maybe Dagger. CUDA/AMDGPU already do a lot of ugly stuff with the source and I expect complications.

In the case of CUDA/AMDGPU, my idea is to instrumentalize kernel launch and data movements, which Extrae should already be capable of detect.

[clasqui]

For the PerMedCoE use case, I agree with targeting Distributed and its variants. I would also try the Base.Threads, as it seems to be the basic shared memory model in Julia.